March 20, 2025 • 47 mins
In this episode of Inside Cancer Careers, Dr. Tony Kerlavage and Jeff Schilling discuss the role of AI in cancer research and the challenges and safety measures associated with its use. They emphasize the need for standardized data and the importance of addressing biases in AI models. They also discuss the recruitment of talent in data science and the potential for partnerships with Silicon Valley. They highlight the complexity of AI and the need for continuous learning in the field. The episode concludes with advice for aspiring data scientists.
SHOW NOTES
Tony Kerlavage, Ph.D.
Jeff Shilling, M.S.
Center for Biomedical Informatics and Information Technology (CBIIT)
Cancer Data Science Blog: Growing the Field - NCI Fellowship Opportunities in Data Science
National Academy of Medicine: Generative AI & LLMs in Health & Medicine Conference
ChatGPT
GovCIO Healthcast Podcast: 50 Years of Cancer: Tech's Role in Fighting Cancer
ARPA-H
Episode 8: Computation to Improve Therapy & Finances with Dr. Peng Jiang
AD: NCI Cancer AI Conversations Seminar Series
YOUR TURN RECOMMENDATIONS
Braiding Sweetgrass: Indigenous Wisdom, Scientific Knowledge and the Teachings of Plants by Robin Wall Kimmerer
Golf Beneath the Surface: The New Science of Golf Psychology by Raymond Prior, Ph.D.
The Diamond Age Or A Young Lady's Illustrated Primer by Neal Stephenson
CENTER FOR CANCER TRAINING (CCT)
Inside Cancer Careers Podcast
CCT Website
LEARN MORE FROM THE NATIONAL CANCER INSTITUTE
Online
By Phone: 1-800-4-CANCER (1-800-422-6237)
U.S. Department of Health and Human Services
National Institutes of Health
National Cancer Institute
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
OLIVER BOGLER (00:00):
This week we are bringing you a repeat of our first episode from 2024 – one of
our most popular and on the topic of AI which of course continues to move at a blistering
pace. One of the guests – Dr. Kerlavage - has since retired, but our conversation with
Jeff Schilling, NCI’s CIO, remains relevant and interesting. Please enjoy the episode.
(00:27):
Hello and welcome to Inside Cancer Careers,
a podcast from the National Cancer Institute. I'm your host,
Oliver Bogler. I work at the NCI in the Center for Cancer Training. On Inside Cancer Careers,
we explore all the different ways that people join the fight against disease and hear their stories.
This is our first episode of 2024 and we are changing things up a little. We've heard from our
(00:48):
listeners and we will be making shorter episodes while still giving you a clear view of the
many interesting things our guests do in cancer research and the career paths that they are on.
Today, we're talking to two leaders at NCI about
AI in cancer research and what the future of data science looks like.
Listen through to the end of the show to hear our guests make some interesting
(01:08):
recommendations and where we invite you to take your turn.
It's a pleasure to welcome Dr. Tony Kerlavage,
Director of NCI's Center for Biomedical Informatics and Information Technology. Welcome.
TONY KERLAVAGE (01:20):
Thank you, Oliver. Pleasure to be here.
OLIVER BOGLER (01:22):
Welcome also to Jeff Schilling, NCI's Chief Information Officer.
JEFF SHILLING (01:26):
Well, thank you. Excited to be here.
OLIVER BOGLER (01:29):
Tony, a year ago, you wrote a blog post, Growing the Field,
NCI Fellowship Opportunities in Data Science, in which you describe the
rapidly growing field of data science and the opportunities for joining the field
with a cancer focus at the NCI. What's changed in the intervening 12 months?
TONY KERLAVAGE (01:48):
Well, Oliver, I mean, one thing I'll say is you can't open an issue of
the NCI news briefing, an informatics smart brief, the Washington Post technology 202
without seeing multiple articles on artificial intelligence. This is a rapidly developing field
and the articles aren't always consistent as to whether AI is a good thing or a bad
(02:10):
thing. Having tremendous breakthroughs or isn't quite ready for prime time.
There's a lot of mixed feelings about AI right now. We know that it has tremendous potential,
but there was a recent National Academy of Medicine conference on AI. And the
participants there were very excited about AI, which holds a lot of potential. But
(02:33):
there’s also acknowledgment that it needs to be consistently monitored to assure that
it keeps performing. So, it clearly is the direction that a lot of work
is going and can really change the face of what can happen in cancer research.
OLIVER BOGLER (02:51):
So you also pointed in that article to the fact that there's a strong
demand for people to join data science. Is AI accelerating that even further?
TONY KERLAVAGE (03:01):
I think so. I think there's certainly a lot of buzz around AI,
and it's certainly out in the lay community as well. But certainly a lot of researchers
are looking at this or are, you know, experimenting with the tools that are
available to them. I know a lot of my colleagues have been, you know, they've been saying, hey,
(03:22):
how can we get access to this within the NCI? Can we get a ChatGPT set up?
I know Jeff can address that issue, but everybody wants to try out these new tools and see how they
can apply them and really radically change how they're doing their data science.
OLIVER BOGLER (03:40):
So Jeff, Tony already mentioned ChatGPT. It burst on the scene about a year
or so ago. And the world has been abuzz with artificial intelligence, great expectations,
and serious concerns have been swamping the media, along with quite a lot of
human drama. AI has been part of the cancer research effort for a while, though, right?
JEFF SHILLING (03:59):
Yes, definitely. I would say it's this generative capability that
really shocked everyone. The amazing speed and thoroughness of the interaction with
these new large language models that when it happened, it was just over a year ago,
(04:19):
maybe a year and a week ago. So of course, AI, though, one of the flavors of AI,
since there's, I think, considered seven or eight different flavors of AI.
I think the biggest one that we've been using in cancer is the one on
images. So we've been able to take a lot of radiology images, a lot of scientific images,
(04:41):
it's like pure data. So there's very little human bias or significance or subtlety that you have to
worry about. You're just basically saying, I know these patients have had this outcome.
Here are these images that we took five years ago. Can you find a difference in those things?
So, it's great for doing work like that, that we're doing on the research side.
(05:05):
And certainly for all the diagnostics that's happening to help with the speed
that radiologists can read images and go through images and keep the right ones,
annotate the right ones. So there's a lot of that, that was happening in industry for 20 years,
right? You have cameras taking very quick photos and things like that.
(05:25):
But I will say that has all been a slow growth where you can feel in your mind that you have
a grasp on how it's going to impact the NCI or the cancer community in general. But when
DALL-E and the other image generation ones came out and now the large language models,
(05:46):
everybody, to me, feels a little nervous, almost like tipped over. Where you're like,
wow, this is so powerful that we're missing two big things. We're missing really, really good
data to make sure that what we feed it is going to come out and be controlled. And the same thing
I said that we have with images, we don't really have with a lot of the textual data that we have.
(06:10):
The other thing is we really don't have a construct to control it, to say what is valid,
what is accurate. And we have such a hard lesson learned with social media coming out,
all of us thinking it's great. We all have Twitter accounts. We all have these social
media accounts. And all of a sudden it starts to be used in a way that
no one predicted. And all of a sudden we have no control. So I think now with AI people are saying,
(06:34):
this is even more dangerous. And we need to get some of these controls in place. So.
TONY KERLAVAGE (06:40):
Yeah, if I could just add onto that, one of the things that has jumped out to
me is that these ChatGPT tools will do whatever you ask them to do. So people have demonstrated
that they can generate misinformation. They can generate deep fake images,
for example. So this is part of, I think, the challenge that we have with the
general public. And sometimes people are skeptical about science to start with. So
(07:04):
I think it's incumbent upon us to really demonstrate that we can control how we use
these tools and use them in a manner that's not going to create misinformation or in some way
compromise people's privacy. I think those are the big issues that people are concerned about.
OLIVER BOGLER (07:22):
So what is the NCI doing in order to allow its scientists and
perhaps cancer researchers across the country to access this technology, but do it in a safe way?
JEFF SHILLING (07:32):
I think we're just getting started. I don't really have a great answer. We
have the standard system controls that we use, which are basically saying the computational
capacity that we use must be secure. But we just talked about something completely different,
which is the output of that can be wrong or false or even, and maybe not accidentally so,
(08:00):
but purposely so. And so we don't have that, really, control in any model yet. So I think
that's something that we're going to have to start having a lot of discussion on. And again,
we take it for granted. We've been at the National Cancer Institute
for a long time. We take it for granted that what the NCI says is taken as fact.
(08:24):
And we saw with COVID and the vaccines, and all of a sudden people just were like, well,
I don't believe that. And so I think we don't have those things in place.
We don't have like a seal of approval or a UL listing, let's say, something like
that. Underwriters Laboratories, you know, we don't have an independent body saying,
we validated this and it's safe. So I think we're just going to have to start doing that.
(08:48):
But we do know that data we have to put into these models is not yet good enough.
We do have a lot of work trying to shore up that data quality in a systemic way.
TONY KERLAVAGE (09:01):
Yeah, if it's OK, Oliver, I'd like to add on to two things that Jeff brought up,
two very important things. Certainly in terms of having confidence in systems themselves,
Jeff's absolutely right. We don't have this rigorous testing of things yet.
But there are a couple of big things that we know where we need to start.
(09:23):
For example, we know we can't just take this large language model,
the ChatGPT-4 type of model that has all of the data on the internet flowing into it. We need
to have private models. We need to control what information is going into it. And that
can really help cut down on misinformation or the hallucinations that everybody reads about.
(09:47):
We know the things that are required for these systems at a very high level. Implementing it is
another issue to Jeff's point, but we know it needs to be open architecture, that it
has to have, you know, well-documented APIs to interact with the system, clear and thorough
documentation of the whole system, and a clear understanding of how the systems can be used.
(10:09):
The second part, Jeff brought up - very important - about the data itself. Again, there's been a
lot of conversation around this, and there are a handful or so different very important points
about the data. First of all, to be AI ready. A study has to be properly designed, right? So you
(10:30):
need to, for example, make sure that there's no bias in your data sets. They must be a sufficient
sample size. They've got to be large enough that these very powerful models are actually
going to give some results without having certain biases already built in from the algorithm itself.
OLIVER BOGLER (10:48):
And let me just ask on the biases,
is that, for example, a non-representative group of samples coming into the data set?
TONY KERLAVAGE (10:55):
Yeah, that would be one of them, absolutely. Representation across
various populations is a key part of that study design right up front. So I would
say that a lot of the data sets, a lot of the legacy data sets we have right now,
are going to be very challenging, to use those in an AI context. And I think we need to start
thinking prospectively about designing the experiments in such a way that we weed out
(11:21):
that bias, have the proper representation across populations, have sufficient sample
size and completeness of the data. Missing data is another pitfall with these systems.
We also have to make sure that you know identifiable information is protected,
that the data themselves are shared in a fair manner, and quality is key. Consistency in format,
(11:46):
and the nomenclature used, and the terminologies are being used, and things being properly
formatted. If that's not there, I think you can end up confusing these systems.
OLIVER BOGLER (11:57):
So I think one of the things I struggle with as a non-expert who is very
interested in AI is this difference that you've already alluded to between sort of
what has gone on in the past, for example, image analysis, right? So if I see a lecture
on an AI ML approach to image analysis, I can understand that an image is taken,
vectors are created, for example, around edges or something like that, and then this is classified,
(12:22):
and then a model is trained to distinguish two populations and then test it with some
sample data that comes in. I get all that. And I think, although I don't understand it
mathematically, I sort of intuitively understand that if I was smart enough,
I could. But when you talk about generative AI, you know, I listened to a lot of podcasts
(12:42):
about this. I hear experts say no one can really explain how it works. Is that true?
TONY KERLAVAGE (12:48):
I think it's very hard to understand how it works. It's, you know,
it does seem very much like a black box. Clearly you can define these
things mathematically. People smarter than me can do that. But for most of us,
I think it really is a black box and there's a big level of trust involved in it.
OLIVER BOGLER (13:07):
Still, it's an exciting time. So let me take a slight turn and
sort of ask you about recruiting talent to the NCI. You're both in leadership positions
here. Your fields are adjacent to Silicon Valley. That's obviously a great place to
go if you're interested in data science, if you're interested in AI. How do we at the
NCI and in the cancer field in general, how do we compete for the talent that we need?
TONY KERLAVAGE (13:31):
Well, competing with Silicon Valley is a really tall order. We do have
constraints, certainly in the government and even in academia, about getting really talented people.
I think we get people who are motivated by things other than the very slick Silicon Valley jobs and
(13:56):
exorbitant salaries. People have a passion for the research, passion for the work that is being done.
But I think there are opportunities to take a tool like this... There are some people who really love
working in the space and working on algorithms, that sort of thing, which is great. At the NCI,
I think it's more about the people who work here in the Intramural Program or the people
(14:18):
that we're funding really are more of an applied science, taking these tools and trying to solve
a biomedical problem with them. Trying to do better predictions for cancer outcomes,
for example. So I think those are the people who have a very different
background we're looking to recruit and would be attracted to coming to the NCI.
JEFF SHILLING (14:41):
I think Oliver too, if you're looking at young people,
you know, so people basically in that transition period,
determining where they're going to take a position. One of the things Tony and I
discussed was creating some starting … starter positions related to a clear career path,
(15:03):
You know, in preparation for this, I wrote some notes about the kind of person that would be
successful. And really, I think it's someone who probably underestimates the knowledge that
they have and the skills that they have. They don't focus on that. They focus more on the
vastness of what they don't know. And how could I possibly get into data science? I'm a bench
(15:28):
researcher or I've only worked with agents. Like, well, the reality is all this is new
all the time. And you really need, to me, if you can trust that what you know is very valuable,
and then if you can add that your interest in data science or information technology or technology,
(15:48):
kind of any kind, and then you really look inside yourself and say, I have a continuous learning
model. I like to learn. Well, that's probably the number one key to success in this space.
And then I think the other thing you really have to want to do is connect
the scientists to this technology. And that's really what Tony and I are doing
(16:11):
all the time. We're kind of reading a whole bunch of things, understanding,
trying to understand in a way that we can translate it to the scientists who
will ask what we think are going to be very complicated questions. But in the end are very,
very simple questions. How do I get access to this large language model so I can play around and see
(16:31):
what it could do? And then all of a sudden, it doesn't really become a very complicated question.
So I think it's really that. It's to us, we can be very attractive,
but we're going to have to either catch people in a career transition,
like I wanna get out of the scientific space I'm in a little bit and go into this other space,
(16:52):
or start something that's a little more advantageous to somebody just coming out
of graduate school and not two postdocs later, right? Not five, six years later.
OLIVER BOGLER (17:01):
CBIIT has created training programs for people like that, right?
JEFF SHILLING (17:06):
We have, but it has been a very, not meant to recruit people into our space. It's,
it's kind of been an opportunity for them to learn about what is,
what is data science and what does it mean?
OLIVER BOGLER (17:19):
So I'll just note that we interviewed Dr. Peng Jiang from the
intramural program a few months ago, and he had an interesting story. He graduated from Princeton
with a computer science degree and initially went to work for Pixar, and then for personal
reasons became motivated to work on cancer and is now an independent investigator at the NCI.
(17:42):
Jeff, I wanted to follow up with you on another strategy that I think you've told me about, which
is that forming partnerships with some of the biggest entities in Silicon Valley is another way
to bring both their technology and opportunities perhaps for people to work in the cancer space and
not be disconnected from Silicon Valley, in fact, to have a connection there. Isn't that right?
JEFF SHILLING (18:03):
Well, this is where I would say we utilize the power of the
National Cancer Institute. We are very large. We have a lot of work going on.
We have a lot of people doing that work. And it's very enticing to these players in this
space to be able to work with us because they can. They don't have to train us.
(18:28):
I would say, I don't know how replicable that is to other scientific groups. Because we try to just
use what we can and it can happen quickly. Oliver, you've seen it where the vendors,
when we ask for a call, they get on the line. When we say we want to do this, they're like,
how can we help? I don't think that's everyone's experience. I think other people are going
(18:53):
to have to use other capabilities to get these players to play with them.
But it's also possible to partner with us and work with them. So I would say the biggest thing is
realizing that these vendors desperately need us. They need our vision, our technical understanding
(19:13):
of the space, and they need the data, and the people to work with. And we're also so
research-focused that we don't have just this clinical space that when they go to maybe work
at a cancer center or university, and all of a sudden, it's a very different kind of agenda. So,
and we know they have a lot of that going on. So I think that's been helpful for us.
TONY KERLAVAGE (19:38):
I think it's actually unavoidable working with these large companies,
right? Although, you know, NIH has a culture of open science, of open data, of open software,
and I don't think that's changed necessarily. You know, we fund academicians to develop tools,
but the field is just advanced. The IT field has advanced so rapidly. Whether
(20:04):
it’s just large enterprise systems or AI, as we've been talking about this,
this sort of very emerging thing right now with the large language models etc.
There's no way that academia can actually keep up with it. Academicians, of course,
are constantly contributing to this and a lot of stuff that the commercial companies build
are based upon things that are actually discovered in academic environments.
(20:28):
But for us, we can't afford to do all of this on our own. We have to take advantage
of all of this investment that these large companies are making in infrastructure and
then offering to the world. And to Jeff's point, they love working with us because
what they can learn from what we're doing can be used in a much wider context. But I
(20:54):
think the work that we do will be accelerated by just taking some of these off the shelf
platforms that they have and implementing them in the context of our own research.
OLIVER BOGLER (21:06):
So someone who comes to the NCI to do data science, to do AI,
will have the opportunity to be connected with these efforts and these other big players.
TONY KERLAVAGE (21:16):
Absolutely, absolutely.
OLIVER BOGLER (21:17):
So you've already talked about data and how critical it is to have
data. And of course, the NCI has several sort of marquee projects in its intramural
program where large data sets are being collected. But Jeff, a couple of years ago,
when you and Tony were on a pod for GovCIO, you talked about how in the cancer centers,
the 70 plus cancer centers that NCI designates, there's a lot of data,
(21:40):
but it's all kind of in its own little silos. How is the NCI tackling this data landscape?
JEFF SHILLING (21:48):
Well, I'm going to transfer this one to Tony, because I think this is really,
he's going to really give us some really thoughtful answers for this.
TONY KERLAVAGE (21:54):
Thanks, Jeff. That is a huge problem. And we've been having conversations
with a lot of the folks out at those cancer centers. They've come to the NCI and are saying,
we know this is a problem. We need your help in solving this. There have been a couple of places
that are actually really focusing a lot of energy into doing this as well within their own systems.
(22:18):
But the real problem is that their clinical research centers or their clinical centers
where they're doing treatment are completely disconnected from
their research centers even within the same institution. And that's by design
because electronic health record systems, for example, are designed to be insular,
to protect patient privacy and other issues. And those systems are very important for
(22:43):
billing and other things like that. But the information that they contain can be
extremely valuable on the research side of the house. So they struggle with this internally.
I would just say that it's a very hard problem to solve, but I'm optimistic that some of the
new projects that are going to be coming in this new ARPA-H program, where we are engaged with them
(23:05):
on this biomedical data fabric project, and a number of the activities that we've laid out,
the different technical areas in there, are addressing exactly this issue of getting all
of these data together, being able to utilize them for whatever purpose is needed, whether it's
in clinical treatment or in basic research. And I think we're just gonna have to wait and see what's
(23:30):
actually gonna come out of these, but the ARPA-H model is one that actually fosters innovation. I
think we're gonna see some radical new things being tried there. I think there are gonna be
a lot of opportunities just to try to tackle this problem in a number of different ways.
So I'm optimistic we're gonna see some real breakthroughs over the next couple of years or so.
JEFF SHILLING (23:51):
Tony said it earlier, we have to start collecting data with the understanding
that it's going to be used in the long term, not just for its initial intended purpose. And that's,
I think, the big change that has to happen in the next decade. All the data is, we can,
is going to be repurposed. So it's collected in a standard way, it's ordered according to some
(24:15):
ontology that we all agree upon, and then we're all working on moving that structure forward. So
it's not too hard to utilize initially and it's helpful later on. Right now, Tony and I, we're
just seeing this giant morass of data that's been collected in the past. And really,
(24:35):
are we really gonna be able to clean that data up and then say it's accurate? That's a tall order.
OLIVER BOGLER (24:41):
It strikes me a little bit like the data standards that, for example,
international finance uses, which means you can take a credit card from the
United States and go visit your, you know, someone in Europe and use the credit card,
right? It all works because they've got interchangeability of data, right?
TONY KERLAVAGE (24:58):
Yeah, absolutely. Of course, they have a much simpler problem to solve, right?
OLIVER BOGLER (25:02):
I understand.
TONY KERLAVAGE (25:02):
They have a very small number of variables that they need to deal with. And what
we're dealing with all of biology here. And that's really the challenge. And I believe, that is the
biggest problem, is coming to some standardized terminologies, some standardized data elements.
(25:22):
This is going to be a big part of what I was talking about in terms of these ARPA-H projects.
And I, this is the problem we also have with all of these legacy data, because these have all
grown up independently, even though we have some centralized resources in our own semantics team
at CBIIT, who do a yeoman's job of making all of these sets of common data elements available, sets
(25:46):
of standard terminologies available. The issue is that every institution has their own way of doing
things, and they may adopt these things, they may add to them and modify them so that this just
grows and grows and grows. That's the key problem we need to solve. And hopefully this is something
that artificial intelligence may start to help us with, at least with the legacy data. If we can
(26:11):
feed it a lot of this information and some cues about how to map these things together, that's
one of the areas I think that's gonna be … benefit from, one of these technical areas with the ARPA-H
program because it's focused on exactly that. How can we leverage these language models to help
harmonize data, which is a task that takes people months and months and months to do manually today?
OLIVER BOGLER (26:36):
Yeah, and of course you have to future proof it too,
because cancer research is not standing still. New markers, new genes, new treatments,
new prevention strategies are being developed all the time. So it sounds to me like it's
really phenomenally complex, but also incredibly exciting and a great time to be in this field.
TONY KERLAVAGE (26:54):
It is a very exciting time and I think it's going to be terrific for
people trying to get into the field now. I think it's yet another inflection point
for people's careers because there are all of these wonderful tools available now. You know,
things that, you know, when I started back doing bioinformatics decades ago,
(27:15):
there weren't all of these tools. We built all of our own tools. And now we've got these suites
of tools available to us. I find it a very, very exciting time for young investigators.
OLIVER BOGLER (27:25):
On that note, let's take a quick break. And when we come back, we'll
talk about career paths and how you can get into this exciting field just like Tony and Jeff did.
[Music]
JULI KLEMM (27:37):
NCI is going to be launching a new series that we're
calling Cancer AI Conversations. And this is going to be a regular series
that explores emerging topics at the intersection of AI and cancer research.
SEAN HANLON (27:51):
We're going to bring together
experts to share some examples, ideas around emerging AI topics.
Jennifer CouchAnd the idea of
these conversations is to have a series of short presentations followed by a
conversation between some experts with some different viewpoints on different AI topics.
JULI KLEMM (28:10):
AI is obviously a huge topic right now with a lot of excitement and
concern and confusion, but the NCI is committed to advancing the appropriate use of AI for good.
SEAN HANLON (28:20):
We'll have a moderated discussion so we can really get into some of these topics
that we don't have time to build a whole workshop around. But we want to hear them
as they're emerging in real time and are really helping to move the AI field forward.
JULI KLEMM (28:35):
So this isn't a standard webinar. Our goal is something more casual
and conversational. Like, eavesdropping on a really interesting conversation.
Jennifer CouchThis is for everybody,
but especially for cancer researchers and those that are interested in where cancer
AI is going and where it's going to be going in the future of cancer research.
SEAN HANLON (28:54):
Experts will get more out of the details and some of the technical aspects
while the more cancer biology or clinical focus people will learn more about some
of the emerging areas of Cancer AI that are likely to impact the field moving forward.
JULI KLEMM (29:11):
So we're launching Cancer AI Conversations in January,
and our first episode will be January 23rd at 11 a.m. And our first topic is
gonna be on understanding the role of prompt engineering in generative AI.
SEAN HANLON (29:25):
We'll have three presenters, Hoifung Poon from Microsoft Research,
Claus Wilke from University of Texas at Austin and Alexander Johansen from Stanford University.
JULI KLEMM (29:38):
Prompt engineering is the emerging discipline of designing inputs for generative
AI tools that'll produce optimal outputs. We hope you'll join us, but if you can't,
we will be recording these and posting them on the NCI website so that you can catch them later.
JENNIFER COUCH (29:53):
We'll put a link in the show notes.
[Music ends]
OLIVER BOGLER:
And we're back. In our podcast, we're always interested in the career paths of people engaged
in the fight against cancer and what motivates them. Jeff, let's start at the beginning. You
(30:13):
studied biochemistry and molecular biology in college. What got you started in science?
JEFF SHILLING (30:18):
Well, I think all paths are a mixture of good and bad. I would say for me,
I had a spouse who was in a PhD program, and she was very motivational to me to really think about
(30:38):
what I wanted to do. And in the laboratory, I was working on a very exciting project with
chaperonins and our work was very successful and resulted in a Nature publication. And that was a
lot of years, you know, several years of work in the cold room. And I was so disappointed with how,
(31:05):
you know, as working in this scientific project that was the pinnacle of our work. That was
it. We were never going to be greater than that moment. And it just wasn't great enough for me.
And the other thing I think was the speed at which … science is, it goes very slowly. You have to be
focused so much. And I had the opportunity then to, you know, the university was just switching
(31:32):
over to the internet, to a fiber optic internet connection. And so they needed people to help
with their departments. So I took the training and I gave a little bit of a tutorial to the
department. Department of Physiological Chemistry at the University of Wisconsin. And the chair of
the department was there. And afterwards he said, "Jeff, you really have a knack for this." Now I
(31:56):
had never talked to that person before. So for him to approach me, that really gave me a lot
of confidence to kind of make that transition. So I think today you can live in both worlds a lot
easier than then, there was no person doing IT, there was no central IT, there was no Internet.
So I would say you have to, the door opens for you. And it might open many times, but you have
(32:25):
to walk through it. You have to take that chance. I think it's a lot, you know, it is hard today,
but it's a lot easier to go back and forth. You wanted to go back in the laboratory. We
have several staff in our operation that were, they’re PhDs, they were in the lab. They made that
transition and it's very exciting because now they can work on hundreds of projects, not just one.
(32:46):
So if, if that's your knack to be in that, that middle ground and to want to do a lot of things,
juggle a lot of balls. Uh, I think that's really, um, uh, kind of what's needed, that mindset.
OLIVER BOGLER (33:00):
Tony, how about you? You started in chemistry, right?
TONY KERLAVAGE (33:04):
Well, I find it's quite fascinating that Jeff and I have had
such a parallel experience here. I, yeah, I started in, I have a, degrees in chemistry,
but I was actually doing biochemistry. You know, I was at University of California San
Diego in a chemistry department only because they didn't have a biochemistry department at
the time. They do now, of course. But yeah, same sort of thing in the, in the cold room doing,
(33:30):
doing protein characterization. Protein sequencing actually, when the very,
very first protein sequencers came out. This was back in the early 80s, late 70s, early 80s.
And then had the opportunity, continued that sort of work through a postdoc
(33:50):
and then came to the NIH and started working on neurotransmitter receptors,
same sort of thing. A lab full of biochemists and pharmacologists trying
to understand neurotransmitter receptors. Then a door opened. Back to Jeff's point,
this is so important about doors opening and going through them. The first neurotransmitter was,
the gene for it was sequenced. And that changed everything. We dropped, we all dropped everything
(34:15):
we were doing. We taught ourselves molecular biology and we started fishing for genes.
And, we also had the opportunity to get one of the very first automated DNA sequencers
from Applied Biosystems at the time. And so we started doing automated DNA.
We had done a lot of sequencing on gels and that sort of stuff, but when that came out,
it really changed everything. And the amount of data we started generating,
(34:38):
of course, increased tremendously too. So we had to, now all of a sudden, what are we
doing with these data? How do we assemble these data together? How do we analyze what they are?
I guess I had an interest in that at the time,
so I basically taught myself how to use some of these tools. I'd never used them before.
Using VAX computers at the time, where these packages were installed, cobbling some things
(35:05):
together and really starting to manage this information. And then over quite a few years and
leaving the NIH going out into the nonprofit world and then into the commercial sector,
working on sequencing the human genome. And it just exploded over time. I started by pulling
cables through the ceiling because we had to connect computers together to manage all of these
(35:28):
data from the sequencers and then had a group of folks who were starting to develop software tools.
Things didn't exist, so we had to develop them from scratch. This was right around the
time the internet was just becoming, you know, something that we could use as another tool,
too. And so it just evolved over time, and I had the wonderful opportunity almost 13 years ago now
(35:49):
to come back to the NIH, this time at the NCI, and now apply all of that life experience from,
you know, biochemistry to molecular biology to IT and informatics into this role at the NCI. So
it's been a number of doors that have opened at very opportune times. And I just sort of,
(36:12):
you know, I looked for the open doors and I ran through them.
OLIVER BOGLER (36:16):
So you both talked about seeking a way to make greater impact and also just staying
with a field, watching a field that's exploding and evolving and just being part of that. Jeff,
Tony told us how he got back to NCI. I wonder if
you could complete your story. What led you to the current role that you're in?
JEFF SHILLING (36:35):
Well, you know, as Tony was speaking, I thought I wanted to make sure
to communicate (36:40):
there's no clear path to what we're doing that I understand.
I think maybe you get a degree in data science or, you know, you could certainly
get degrees in information technology today, but what we have is this mixture.
I wanted to say this going backwards a little bit is, so I was in the lab and I needed to get a job
(37:02):
in IT. Well, I don't have any IT experience at all other than just doing a few computers in the lab.
I called this woman, Lisa Longenecker, and every day, about this open position, for two months till
she finally hired me half time. And then I had to find another half time job. So for two years,
(37:24):
I worked half time in two departments just to learn. And I was so focused, like Tony said,
you're jumping at every opportunity to learn because I didn't have any formal training. So
then my wife took a postdoc at National Institute of Environmental Health Sciences and I was able
to get a job there. And then after a few years switched from just doing IT support to doing
(37:50):
more IT programming and general IT work at the National Toxicology Program, which is associated
with NIEHS as well. So then a couple of years, we went to Lawrence Livermore Laboratory. And then
my wife and I were both recruited here in the intramural program, Center for Cancer Research.
(38:11):
I remember very much because my wife was interviewing for a job and I just happened
to go with her. So we're at dinner. And we knew the director had worked at NIEHS,
Carl Barrett. And as he was interviewing her, he said, Jeff, what do you do? And I said,
I do this. And he's like, oh my God, I need you more than Michelle. So really,
again, you can call it a door. You can call it whatever. But I think the vacuum of people
(38:37):
who do what we do is still great. If you just get these skills, just like Tony said,
he's pulling wire. He doesn't have to do that. He could have sat back and said,
we can't do this. We don't have any wire. He's like, well, how hard can wire be to pull?
He probably got yelled at. I guarantee he got yelled at for doing that. But we need to do
(38:58):
this. The show must go on. Say however you want to say it. So really, that was it. I think it was
having the skill set, being able to articulate how you can connect the science and the IT or
the informatics, how you can do that. And build the relationships along the way. You have to have
these relationships with the scientists to be able to, I think, get some of these positions.
OLIVER BOGLER (39:23):
Sounds like, almost like a startup ethos, right? That you have to bring to this,
to work in large complex organizations.
JEFF SHILLING (39:30):
Yeah, I wouldn't disagree. I can only say I've never done a startup,
unfortunately. It sounds like fun.
OLIVER BOGLER (39:37):
So in closing then, I wonder what advice you might have to our listeners,
people who are interested in data science and cancer research. Obviously they may have some
more opportunities that didn't exist when you were at their stage in terms of structured
training opportunities. But Tony, let's start with you. What would you say to them?
TONY KERLAVAGE (39:55):
I, you know, again, building on, you know,
both the path that Jeff and I both took, the take home lesson, I think, is flexibility,
is being open to possibilities. You never know what path a career is going to take. I think
back to when I was working at the Institute for Genomic Research, and we were hiring,
(40:20):
you know, it wasn't just molecular biologists and biochemists, we were hiring physicists.
We're hiring people from various walks of life coming together with different training,
different backgrounds and skills coming together to solve a big problem. So my advice would be,
don't think that you don't have the right background, the right training to get into
(40:42):
something that you might have a passion for. Just take that leap of faith to go
through that door and it could be a wonderful opportunity for you that you never even imagined.
OLIVER BOGLER (40:54):
Jeff?
JEFF SHILLING (40:55):
I would say one of my sayings is, when you're an expert, everything you know is old.
No one knows the future. No one knows tomorrow. There is an opportunity for you to learn that.
You have to understand that when you go into a new field, you're going to be overconfident
just because you read a few things that you know what's going on. But the reality is the
(41:19):
situation is very complex. That's why they need people to help figure it out. And so,
you know, just going there, humble, understand that what you know from your other work is very,
very valuable and that learning these new fields is - everyone's in the same
boat. No one knows what's going on. We’re all trying to stay afloat and continuously learn.
(41:43):
And you will one day be the expert, and everything you know will be old.
OLIVER BOGLER (41:48):
Well, thank you both very much for this very
interesting conversation and sharing your advice.
TONY KERLAVAGE (41:53):
You're quite welcome, it's been a pleasure.
JEFF SHILLING (41:56):
Thank you, Oliver.
Oliver Bogler
Now it's time for a segment we call your turn because it's a chance for
our listeners to send in recommendations that they would like to share. If you're listening,
then you're invited to take your turn. Send us a tip for a book, a video, a podcast,
or a talk that you found inspirational, amusing, interesting. You can send these to us
(42:20):
at NCIICC@nih.gov. Record a voice memo and send it along. We may just play it in an upcoming episode.
Now I'd like to invite our guests to take their turn. Let's start with you, Tony.
TONY KERLAVAGE (42:34):
Oliver, there's a book I read recently that I found very impactful. And the
book is called Braiding Sweetgrass by Robin Wall Kimmerer. The subtitle of the book is Indigenous
Wisdom, Scientific Knowledge, and the Teaching of Plants. Robin Wall Kimmerer is a botanist of
Native American heritage and a mother just trying to pass on wisdom to her children.
(43:00):
What I found most fascinating about this book is that what came through was really a new way
of thinking about nature and our place in it. And I highly recommend that book for anyone.
OLIVER BOGLER (43:12):
That sounds interesting, I'm gonna put that on my reading list, thanks. Jeff?
JEFF SHILLING (43:16):
Well, you know, I've always been somewhat of an athlete. I don't really
necessarily always like to watch sports, but I do like to participate in it. And as one gets older,
it's a little harder to do things. So I've, as an adult, I've always really liked to play
golf. And I do read a lot of golf books, both physically and mentally. And one of
the books just came out is called Golf Beneath the Surface: The New Science of Golf Psychology.
(43:44):
And when I started to read it, I was blown away by how very,
very different it was. And it was utilizing all the work that's been done on the brain.
OLIVER BOGLER (43:56):
Hmm, interesting.
JEFF SHILLING (43:57):
And really describing, you know, I was like, this is not a book about
golf at all. It's about how we think and act. How, what do we do when we get afraid? How to
manage our nervousness? What kind of nervousness is good, and is acceptable and how to deal with
that? What kind of nervousness or fear is bad? And that's a sign that we need to change our thinking.
(44:22):
I was so surprised how I'd never heard of these things. Sometimes
I think we have little techniques that we do to manage our life and someone makes
us angry or we're in traffic and we're mad or something like that. I was very,
very surprised and also very excited because all of a sudden people are using all that brain
(44:44):
research that's been happening. And to me, this is kind of the transformation of society to say,
you know, this way we feel, these emotions that we have, that's very, very natural. That is not
something you need to even be concerned about. We were not trained in any of those things, right?
OLIVER BOGLER (45:04):
Sounds useful even if you don't play golf, I guess.
JEFF SHILLING (45:06):
Very useful, yes.
OLIVER BOGLER (45:08):
OK, finally, I'd like to make a recommendation as well. It's a favorite book
of mine that I recently reread. It's called The Diamond Age or A Young Lady's Illustrated Primer
by Neal Stephenson. It was first published in 2000. And I'd remembered it as a book with a
striking vision of how nanotechnology could transform our world when I read it back then.
(45:28):
But this time I enjoyed it much more as a vision of the primer itself. It's a book
that teaches a young girl and imprints on her to provide a bespoke interactive
experience driven by an AI. And it struck me as an incredibly interesting imagining
of what interactive learning might look like with the AI tech that we've
been talking about today. Remarkable book, I think, and well worth a read.
(45:53):
That’s all we have time for on today’s episode of Inside Cancer
Careers! Thank you for joining and thank you to our guests.
We want to hear from you – your stories, your ideas and your feedback are always
welcome. And you are invited to take your turn and make a recommendation
to share with our listeners. You can reach us at NCIICC@nih.gov.
(46:14):
Inside Cancer Careers is a collaboration between NCI’s Office of Communications and
Public Liaison and the Center for Cancer Training.
It is produced by Angela Jones and Astrid Masfar.
Join us every first and third Thursday of the month when new episodes can be found wherever you
listen – subscribe so you won’t miss an episode. I'm your host Oliver Bogler from the National
(46:37):
Cancer Institute and I look forward to sharing your stories here on Inside Cancer Careers.
If you have questions about cancer or comments about this podcast, email us at NCIinfo@nih.gov
or call us at 800-422-6237. And please be sure to mention Inside Cancer Careers in your query.
(47:00):
We are a production of the U.S. Department of Health and Human Services,
National Institutes of Health, National Cancer Institute. Thanks for listening.
This week we are bringing you a repeat of our first episode from 2024 – one of
our most popular and on the topic of AI which of course continues to move at a blistering
pace. One of the guests – Dr. Kerlavage - has since retired, but our conversation with
Jeff Schilling, NCI’s CIO, remains relevant and interesting. Please enjoy the episode.
(00:27):
Hello and welcome to Inside Cancer Careers,
a podcast from the National Cancer Institute. I'm your host,
Oliver Bogler. I work at the NCI in the Center for Cancer Training. On Inside Cancer Careers,
we explore all the different ways that people join the fight against disease and hear their stories.
This is our first episode of 2024 and we are changing things up a little. We've heard from our
(00:48):
listeners and we will be making shorter episodes while still giving you a clear view of the
many interesting things our guests do in cancer research and the career paths that they are on.
Today, we're talking to two leaders at NCI about
AI in cancer research and what the future of data science looks like.
Listen through to the end of the show to hear our guests make some interesting
(01:08):
recommendations and where we invite you to take your turn.
It's a pleasure to welcome Dr. Tony Kerlavage,
Director of NCI's Center for Biomedical Informatics and Information Technology. Welcome.
TONY KERLAVAGE (01:20):
Thank you, Oliver. Pleasure to be here.
OLIVER BOGLER (01:22):
Welcome also to Jeff Schilling, NCI's Chief Information Officer.
JEFF SHILLING (01:26):
Well, thank you. Excited to be here.
OLIVER BOGLER (01:29):
Tony, a year ago, you wrote a blog post, Growing the Field,
NCI Fellowship Opportunities in Data Science, in which you describe the
rapidly growing field of data science and the opportunities for joining the field
with a cancer focus at the NCI. What's changed in the intervening 12 months?
TONY KERLAVAGE (01:48):
Well, Oliver, I mean, one thing I'll say is you can't open an issue of
the NCI news briefing, an informatics smart brief, the Washington Post technology 202
without seeing multiple articles on artificial intelligence. This is a rapidly developing field
and the articles aren't always consistent as to whether AI is a good thing or a bad
(02:10):
thing. Having tremendous breakthroughs or isn't quite ready for prime time.
There's a lot of mixed feelings about AI right now. We know that it has tremendous potential,
but there was a recent National Academy of Medicine conference on AI. And the
participants there were very excited about AI, which holds a lot of potential. But
(02:33):
there’s also acknowledgment that it needs to be consistently monitored to assure that
it keeps performing. So, it clearly is the direction that a lot of work
is going and can really change the face of what can happen in cancer research.
OLIVER BOGLER (02:51):
So you also pointed in that article to the fact that there's a strong
demand for people to join data science. Is AI accelerating that even further?
TONY KERLAVAGE (03:01):
I think so. I think there's certainly a lot of buzz around AI,
and it's certainly out in the lay community as well. But certainly a lot of researchers
are looking at this or are, you know, experimenting with the tools that are
available to them. I know a lot of my colleagues have been, you know, they've been saying, hey,
(03:22):
how can we get access to this within the NCI? Can we get a ChatGPT set up?
I know Jeff can address that issue, but everybody wants to try out these new tools and see how they
can apply them and really radically change how they're doing their data science.
OLIVER BOGLER (03:40):
So Jeff, Tony already mentioned ChatGPT. It burst on the scene about a year
or so ago. And the world has been abuzz with artificial intelligence, great expectations,
and serious concerns have been swamping the media, along with quite a lot of
human drama. AI has been part of the cancer research effort for a while, though, right?
JEFF SHILLING (03:59):
Yes, definitely. I would say it's this generative capability that
really shocked everyone. The amazing speed and thoroughness of the interaction with
these new large language models that when it happened, it was just over a year ago,
(04:19):
maybe a year and a week ago. So of course, AI, though, one of the flavors of AI,
since there's, I think, considered seven or eight different flavors of AI.
I think the biggest one that we've been using in cancer is the one on
images. So we've been able to take a lot of radiology images, a lot of scientific images,
(04:41):
it's like pure data. So there's very little human bias or significance or subtlety that you have to
worry about. You're just basically saying, I know these patients have had this outcome.
Here are these images that we took five years ago. Can you find a difference in those things?
So, it's great for doing work like that, that we're doing on the research side.
(05:05):
And certainly for all the diagnostics that's happening to help with the speed
that radiologists can read images and go through images and keep the right ones,
annotate the right ones. So there's a lot of that, that was happening in industry for 20 years,
right? You have cameras taking very quick photos and things like that.
(05:25):
But I will say that has all been a slow growth where you can feel in your mind that you have
a grasp on how it's going to impact the NCI or the cancer community in general. But when
DALL-E and the other image generation ones came out and now the large language models,
(05:46):
everybody, to me, feels a little nervous, almost like tipped over. Where you're like,
wow, this is so powerful that we're missing two big things. We're missing really, really good
data to make sure that what we feed it is going to come out and be controlled. And the same thing
I said that we have with images, we don't really have with a lot of the textual data that we have.
(06:10):
The other thing is we really don't have a construct to control it, to say what is valid,
what is accurate. And we have such a hard lesson learned with social media coming out,
all of us thinking it's great. We all have Twitter accounts. We all have these social
media accounts. And all of a sudden it starts to be used in a way that
no one predicted. And all of a sudden we have no control. So I think now with AI people are saying,
(06:34):
this is even more dangerous. And we need to get some of these controls in place. So.
TONY KERLAVAGE (06:40):
Yeah, if I could just add onto that, one of the things that has jumped out to
me is that these ChatGPT tools will do whatever you ask them to do. So people have demonstrated
that they can generate misinformation. They can generate deep fake images,
for example. So this is part of, I think, the challenge that we have with the
general public. And sometimes people are skeptical about science to start with. So
(07:04):
I think it's incumbent upon us to really demonstrate that we can control how we use
these tools and use them in a manner that's not going to create misinformation or in some way
compromise people's privacy. I think those are the big issues that people are concerned about.
OLIVER BOGLER (07:22):
So what is the NCI doing in order to allow its scientists and
perhaps cancer researchers across the country to access this technology, but do it in a safe way?
JEFF SHILLING (07:32):
I think we're just getting started. I don't really have a great answer. We
have the standard system controls that we use, which are basically saying the computational
capacity that we use must be secure. But we just talked about something completely different,
which is the output of that can be wrong or false or even, and maybe not accidentally so,
(08:00):
but purposely so. And so we don't have that, really, control in any model yet. So I think
that's something that we're going to have to start having a lot of discussion on. And again,
we take it for granted. We've been at the National Cancer Institute
for a long time. We take it for granted that what the NCI says is taken as fact.
(08:24):
And we saw with COVID and the vaccines, and all of a sudden people just were like, well,
I don't believe that. And so I think we don't have those things in place.
We don't have like a seal of approval or a UL listing, let's say, something like
that. Underwriters Laboratories, you know, we don't have an independent body saying,
we validated this and it's safe. So I think we're just going to have to start doing that.
(08:48):
But we do know that data we have to put into these models is not yet good enough.
We do have a lot of work trying to shore up that data quality in a systemic way.
TONY KERLAVAGE (09:01):
Yeah, if it's OK, Oliver, I'd like to add on to two things that Jeff brought up,
two very important things. Certainly in terms of having confidence in systems themselves,
Jeff's absolutely right. We don't have this rigorous testing of things yet.
But there are a couple of big things that we know where we need to start.
(09:23):
For example, we know we can't just take this large language model,
the ChatGPT-4 type of model that has all of the data on the internet flowing into it. We need
to have private models. We need to control what information is going into it. And that
can really help cut down on misinformation or the hallucinations that everybody reads about.
(09:47):
We know the things that are required for these systems at a very high level. Implementing it is
another issue to Jeff's point, but we know it needs to be open architecture, that it
has to have, you know, well-documented APIs to interact with the system, clear and thorough
documentation of the whole system, and a clear understanding of how the systems can be used.
(10:09):
The second part, Jeff brought up - very important - about the data itself. Again, there's been a
lot of conversation around this, and there are a handful or so different very important points
about the data. First of all, to be AI ready. A study has to be properly designed, right? So you
(10:30):
need to, for example, make sure that there's no bias in your data sets. They must be a sufficient
sample size. They've got to be large enough that these very powerful models are actually
going to give some results without having certain biases already built in from the algorithm itself.
OLIVER BOGLER (10:48):
And let me just ask on the biases,
is that, for example, a non-representative group of samples coming into the data set?
TONY KERLAVAGE (10:55):
Yeah, that would be one of them, absolutely. Representation across
various populations is a key part of that study design right up front. So I would
say that a lot of the data sets, a lot of the legacy data sets we have right now,
are going to be very challenging, to use those in an AI context. And I think we need to start
thinking prospectively about designing the experiments in such a way that we weed out
(11:21):
that bias, have the proper representation across populations, have sufficient sample
size and completeness of the data. Missing data is another pitfall with these systems.
We also have to make sure that you know identifiable information is protected,
that the data themselves are shared in a fair manner, and quality is key. Consistency in format,
(11:46):
and the nomenclature used, and the terminologies are being used, and things being properly
formatted. If that's not there, I think you can end up confusing these systems.
OLIVER BOGLER (11:57):
So I think one of the things I struggle with as a non-expert who is very
interested in AI is this difference that you've already alluded to between sort of
what has gone on in the past, for example, image analysis, right? So if I see a lecture
on an AI ML approach to image analysis, I can understand that an image is taken,
vectors are created, for example, around edges or something like that, and then this is classified,
(12:22):
and then a model is trained to distinguish two populations and then test it with some
sample data that comes in. I get all that. And I think, although I don't understand it
mathematically, I sort of intuitively understand that if I was smart enough,
I could. But when you talk about generative AI, you know, I listened to a lot of podcasts
(12:42):
about this. I hear experts say no one can really explain how it works. Is that true?
TONY KERLAVAGE (12:48):
I think it's very hard to understand how it works. It's, you know,
it does seem very much like a black box. Clearly you can define these
things mathematically. People smarter than me can do that. But for most of us,
I think it really is a black box and there's a big level of trust involved in it.
OLIVER BOGLER (13:07):
Still, it's an exciting time. So let me take a slight turn and
sort of ask you about recruiting talent to the NCI. You're both in leadership positions
here. Your fields are adjacent to Silicon Valley. That's obviously a great place to
go if you're interested in data science, if you're interested in AI. How do we at the
NCI and in the cancer field in general, how do we compete for the talent that we need?
TONY KERLAVAGE (13:31):
Well, competing with Silicon Valley is a really tall order. We do have
constraints, certainly in the government and even in academia, about getting really talented people.
I think we get people who are motivated by things other than the very slick Silicon Valley jobs and
(13:56):
exorbitant salaries. People have a passion for the research, passion for the work that is being done.
But I think there are opportunities to take a tool like this... There are some people who really love
working in the space and working on algorithms, that sort of thing, which is great. At the NCI,
I think it's more about the people who work here in the Intramural Program or the people
(14:18):
that we're funding really are more of an applied science, taking these tools and trying to solve
a biomedical problem with them. Trying to do better predictions for cancer outcomes,
for example. So I think those are the people who have a very different
background we're looking to recruit and would be attracted to coming to the NCI.
JEFF SHILLING (14:41):
I think Oliver too, if you're looking at young people,
you know, so people basically in that transition period,
determining where they're going to take a position. One of the things Tony and I
discussed was creating some starting … starter positions related to a clear career path,
(15:03):
You know, in preparation for this, I wrote some notes about the kind of person that would be
successful. And really, I think it's someone who probably underestimates the knowledge that
they have and the skills that they have. They don't focus on that. They focus more on the
vastness of what they don't know. And how could I possibly get into data science? I'm a bench
(15:28):
researcher or I've only worked with agents. Like, well, the reality is all this is new
all the time. And you really need, to me, if you can trust that what you know is very valuable,
and then if you can add that your interest in data science or information technology or technology,
(15:48):
kind of any kind, and then you really look inside yourself and say, I have a continuous learning
model. I like to learn. Well, that's probably the number one key to success in this space.
And then I think the other thing you really have to want to do is connect
the scientists to this technology. And that's really what Tony and I are doing
(16:11):
all the time. We're kind of reading a whole bunch of things, understanding,
trying to understand in a way that we can translate it to the scientists who
will ask what we think are going to be very complicated questions. But in the end are very,
very simple questions. How do I get access to this large language model so I can play around and see
(16:31):
what it could do? And then all of a sudden, it doesn't really become a very complicated question.
So I think it's really that. It's to us, we can be very attractive,
but we're going to have to either catch people in a career transition,
like I wanna get out of the scientific space I'm in a little bit and go into this other space,
(16:52):
or start something that's a little more advantageous to somebody just coming out
of graduate school and not two postdocs later, right? Not five, six years later.
OLIVER BOGLER (17:01):
CBIIT has created training programs for people like that, right?
JEFF SHILLING (17:06):
We have, but it has been a very, not meant to recruit people into our space. It's,
it's kind of been an opportunity for them to learn about what is,
what is data science and what does it mean?
OLIVER BOGLER (17:19):
So I'll just note that we interviewed Dr. Peng Jiang from the
intramural program a few months ago, and he had an interesting story. He graduated from Princeton
with a computer science degree and initially went to work for Pixar, and then for personal
reasons became motivated to work on cancer and is now an independent investigator at the NCI.
(17:42):
Jeff, I wanted to follow up with you on another strategy that I think you've told me about, which
is that forming partnerships with some of the biggest entities in Silicon Valley is another way
to bring both their technology and opportunities perhaps for people to work in the cancer space and
not be disconnected from Silicon Valley, in fact, to have a connection there. Isn't that right?
JEFF SHILLING (18:03):
Well, this is where I would say we utilize the power of the
National Cancer Institute. We are very large. We have a lot of work going on.
We have a lot of people doing that work. And it's very enticing to these players in this
space to be able to work with us because they can. They don't have to train us.
(18:28):
I would say, I don't know how replicable that is to other scientific groups. Because we try to just
use what we can and it can happen quickly. Oliver, you've seen it where the vendors,
when we ask for a call, they get on the line. When we say we want to do this, they're like,
how can we help? I don't think that's everyone's experience. I think other people are going
(18:53):
to have to use other capabilities to get these players to play with them.
But it's also possible to partner with us and work with them. So I would say the biggest thing is
realizing that these vendors desperately need us. They need our vision, our technical understanding
(19:13):
of the space, and they need the data, and the people to work with. And we're also so
research-focused that we don't have just this clinical space that when they go to maybe work
at a cancer center or university, and all of a sudden, it's a very different kind of agenda. So,
and we know they have a lot of that going on. So I think that's been helpful for us.
TONY KERLAVAGE (19:38):
I think it's actually unavoidable working with these large companies,
right? Although, you know, NIH has a culture of open science, of open data, of open software,
and I don't think that's changed necessarily. You know, we fund academicians to develop tools,
but the field is just advanced. The IT field has advanced so rapidly. Whether
(20:04):
it’s just large enterprise systems or AI, as we've been talking about this,
this sort of very emerging thing right now with the large language models etc.
There's no way that academia can actually keep up with it. Academicians, of course,
are constantly contributing to this and a lot of stuff that the commercial companies build
are based upon things that are actually discovered in academic environments.
(20:28):
But for us, we can't afford to do all of this on our own. We have to take advantage
of all of this investment that these large companies are making in infrastructure and
then offering to the world. And to Jeff's point, they love working with us because
what they can learn from what we're doing can be used in a much wider context. But I
(20:54):
think the work that we do will be accelerated by just taking some of these off the shelf
platforms that they have and implementing them in the context of our own research.
OLIVER BOGLER (21:06):
So someone who comes to the NCI to do data science, to do AI,
will have the opportunity to be connected with these efforts and these other big players.
TONY KERLAVAGE (21:16):
Absolutely, absolutely.
OLIVER BOGLER (21:17):
So you've already talked about data and how critical it is to have
data. And of course, the NCI has several sort of marquee projects in its intramural
program where large data sets are being collected. But Jeff, a couple of years ago,
when you and Tony were on a pod for GovCIO, you talked about how in the cancer centers,
the 70 plus cancer centers that NCI designates, there's a lot of data,
(21:40):
but it's all kind of in its own little silos. How is the NCI tackling this data landscape?
JEFF SHILLING (21:48):
Well, I'm going to transfer this one to Tony, because I think this is really,
he's going to really give us some really thoughtful answers for this.
TONY KERLAVAGE (21:54):
Thanks, Jeff. That is a huge problem. And we've been having conversations
with a lot of the folks out at those cancer centers. They've come to the NCI and are saying,
we know this is a problem. We need your help in solving this. There have been a couple of places
that are actually really focusing a lot of energy into doing this as well within their own systems.
(22:18):
But the real problem is that their clinical research centers or their clinical centers
where they're doing treatment are completely disconnected from
their research centers even within the same institution. And that's by design
because electronic health record systems, for example, are designed to be insular,
to protect patient privacy and other issues. And those systems are very important for
(22:43):
billing and other things like that. But the information that they contain can be
extremely valuable on the research side of the house. So they struggle with this internally.
I would just say that it's a very hard problem to solve, but I'm optimistic that some of the
new projects that are going to be coming in this new ARPA-H program, where we are engaged with them
(23:05):
on this biomedical data fabric project, and a number of the activities that we've laid out,
the different technical areas in there, are addressing exactly this issue of getting all
of these data together, being able to utilize them for whatever purpose is needed, whether it's
in clinical treatment or in basic research. And I think we're just gonna have to wait and see what's
(23:30):
actually gonna come out of these, but the ARPA-H model is one that actually fosters innovation. I
think we're gonna see some radical new things being tried there. I think there are gonna be
a lot of opportunities just to try to tackle this problem in a number of different ways.
So I'm optimistic we're gonna see some real breakthroughs over the next couple of years or so.
JEFF SHILLING (23:51):
Tony said it earlier, we have to start collecting data with the understanding
that it's going to be used in the long term, not just for its initial intended purpose. And that's,
I think, the big change that has to happen in the next decade. All the data is, we can,
is going to be repurposed. So it's collected in a standard way, it's ordered according to some
(24:15):
ontology that we all agree upon, and then we're all working on moving that structure forward. So
it's not too hard to utilize initially and it's helpful later on. Right now, Tony and I, we're
just seeing this giant morass of data that's been collected in the past. And really,
(24:35):
are we really gonna be able to clean that data up and then say it's accurate? That's a tall order.
OLIVER BOGLER (24:41):
It strikes me a little bit like the data standards that, for example,
international finance uses, which means you can take a credit card from the
United States and go visit your, you know, someone in Europe and use the credit card,
right? It all works because they've got interchangeability of data, right?
TONY KERLAVAGE (24:58):
Yeah, absolutely. Of course, they have a much simpler problem to solve, right?
OLIVER BOGLER (25:02):
I understand.
TONY KERLAVAGE (25:02):
They have a very small number of variables that they need to deal with. And what
we're dealing with all of biology here. And that's really the challenge. And I believe, that is the
biggest problem, is coming to some standardized terminologies, some standardized data elements.
(25:22):
This is going to be a big part of what I was talking about in terms of these ARPA-H projects.
And I, this is the problem we also have with all of these legacy data, because these have all
grown up independently, even though we have some centralized resources in our own semantics team
at CBIIT, who do a yeoman's job of making all of these sets of common data elements available, sets
(25:46):
of standard terminologies available. The issue is that every institution has their own way of doing
things, and they may adopt these things, they may add to them and modify them so that this just
grows and grows and grows. That's the key problem we need to solve. And hopefully this is something
that artificial intelligence may start to help us with, at least with the legacy data. If we can
(26:11):
feed it a lot of this information and some cues about how to map these things together, that's
one of the areas I think that's gonna be … benefit from, one of these technical areas with the ARPA-H
program because it's focused on exactly that. How can we leverage these language models to help
harmonize data, which is a task that takes people months and months and months to do manually today?
OLIVER BOGLER (26:36):
Yeah, and of course you have to future proof it too,
because cancer research is not standing still. New markers, new genes, new treatments,
new prevention strategies are being developed all the time. So it sounds to me like it's
really phenomenally complex, but also incredibly exciting and a great time to be in this field.
TONY KERLAVAGE (26:54):
It is a very exciting time and I think it's going to be terrific for
people trying to get into the field now. I think it's yet another inflection point
for people's careers because there are all of these wonderful tools available now. You know,
things that, you know, when I started back doing bioinformatics decades ago,
(27:15):
there weren't all of these tools. We built all of our own tools. And now we've got these suites
of tools available to us. I find it a very, very exciting time for young investigators.
OLIVER BOGLER (27:25):
On that note, let's take a quick break. And when we come back, we'll
talk about career paths and how you can get into this exciting field just like Tony and Jeff did.
[Music]
JULI KLEMM (27:37):
NCI is going to be launching a new series that we're
calling Cancer AI Conversations. And this is going to be a regular series
that explores emerging topics at the intersection of AI and cancer research.
SEAN HANLON (27:51):
We're going to bring together
experts to share some examples, ideas around emerging AI topics.
Jennifer CouchAnd the idea of
these conversations is to have a series of short presentations followed by a
conversation between some experts with some different viewpoints on different AI topics.
JULI KLEMM (28:10):
AI is obviously a huge topic right now with a lot of excitement and
concern and confusion, but the NCI is committed to advancing the appropriate use of AI for good.
SEAN HANLON (28:20):
We'll have a moderated discussion so we can really get into some of these topics
that we don't have time to build a whole workshop around. But we want to hear them
as they're emerging in real time and are really helping to move the AI field forward.
JULI KLEMM (28:35):
So this isn't a standard webinar. Our goal is something more casual
and conversational. Like, eavesdropping on a really interesting conversation.
Jennifer CouchThis is for everybody,
but especially for cancer researchers and those that are interested in where cancer
AI is going and where it's going to be going in the future of cancer research.
SEAN HANLON (28:54):
Experts will get more out of the details and some of the technical aspects
while the more cancer biology or clinical focus people will learn more about some
of the emerging areas of Cancer AI that are likely to impact the field moving forward.
JULI KLEMM (29:11):
So we're launching Cancer AI Conversations in January,
and our first episode will be January 23rd at 11 a.m. And our first topic is
gonna be on understanding the role of prompt engineering in generative AI.
SEAN HANLON (29:25):
We'll have three presenters, Hoifung Poon from Microsoft Research,
Claus Wilke from University of Texas at Austin and Alexander Johansen from Stanford University.
JULI KLEMM (29:38):
Prompt engineering is the emerging discipline of designing inputs for generative
AI tools that'll produce optimal outputs. We hope you'll join us, but if you can't,
we will be recording these and posting them on the NCI website so that you can catch them later.
JENNIFER COUCH (29:53):
We'll put a link in the show notes.
[Music ends]
OLIVER BOGLER:
And we're back. In our podcast, we're always interested in the career paths of people engaged
in the fight against cancer and what motivates them. Jeff, let's start at the beginning. You
(30:13):
studied biochemistry and molecular biology in college. What got you started in science?
JEFF SHILLING (30:18):
Well, I think all paths are a mixture of good and bad. I would say for me,
I had a spouse who was in a PhD program, and she was very motivational to me to really think about
(30:38):
what I wanted to do. And in the laboratory, I was working on a very exciting project with
chaperonins and our work was very successful and resulted in a Nature publication. And that was a
lot of years, you know, several years of work in the cold room. And I was so disappointed with how,
(31:05):
you know, as working in this scientific project that was the pinnacle of our work. That was
it. We were never going to be greater than that moment. And it just wasn't great enough for me.
And the other thing I think was the speed at which … science is, it goes very slowly. You have to be
focused so much. And I had the opportunity then to, you know, the university was just switching
(31:32):
over to the internet, to a fiber optic internet connection. And so they needed people to help
with their departments. So I took the training and I gave a little bit of a tutorial to the
department. Department of Physiological Chemistry at the University of Wisconsin. And the chair of
the department was there. And afterwards he said, "Jeff, you really have a knack for this." Now I
(31:56):
had never talked to that person before. So for him to approach me, that really gave me a lot
of confidence to kind of make that transition. So I think today you can live in both worlds a lot
easier than then, there was no person doing IT, there was no central IT, there was no Internet.
So I would say you have to, the door opens for you. And it might open many times, but you have
(32:25):
to walk through it. You have to take that chance. I think it's a lot, you know, it is hard today,
but it's a lot easier to go back and forth. You wanted to go back in the laboratory. We
have several staff in our operation that were, they’re PhDs, they were in the lab. They made that
transition and it's very exciting because now they can work on hundreds of projects, not just one.
(32:46):
So if, if that's your knack to be in that, that middle ground and to want to do a lot of things,
juggle a lot of balls. Uh, I think that's really, um, uh, kind of what's needed, that mindset.
OLIVER BOGLER (33:00):
Tony, how about you? You started in chemistry, right?
TONY KERLAVAGE (33:04):
Well, I find it's quite fascinating that Jeff and I have had
such a parallel experience here. I, yeah, I started in, I have a, degrees in chemistry,
but I was actually doing biochemistry. You know, I was at University of California San
Diego in a chemistry department only because they didn't have a biochemistry department at
the time. They do now, of course. But yeah, same sort of thing in the, in the cold room doing,
(33:30):
doing protein characterization. Protein sequencing actually, when the very,
very first protein sequencers came out. This was back in the early 80s, late 70s, early 80s.
And then had the opportunity, continued that sort of work through a postdoc
(33:50):
and then came to the NIH and started working on neurotransmitter receptors,
same sort of thing. A lab full of biochemists and pharmacologists trying
to understand neurotransmitter receptors. Then a door opened. Back to Jeff's point,
this is so important about doors opening and going through them. The first neurotransmitter was,
the gene for it was sequenced. And that changed everything. We dropped, we all dropped everything
(34:15):
we were doing. We taught ourselves molecular biology and we started fishing for genes.
And, we also had the opportunity to get one of the very first automated DNA sequencers
from Applied Biosystems at the time. And so we started doing automated DNA.
We had done a lot of sequencing on gels and that sort of stuff, but when that came out,
it really changed everything. And the amount of data we started generating,
(34:38):
of course, increased tremendously too. So we had to, now all of a sudden, what are we
doing with these data? How do we assemble these data together? How do we analyze what they are?
I guess I had an interest in that at the time,
so I basically taught myself how to use some of these tools. I'd never used them before.
Using VAX computers at the time, where these packages were installed, cobbling some things
(35:05):
together and really starting to manage this information. And then over quite a few years and
leaving the NIH going out into the nonprofit world and then into the commercial sector,
working on sequencing the human genome. And it just exploded over time. I started by pulling
cables through the ceiling because we had to connect computers together to manage all of these
(35:28):
data from the sequencers and then had a group of folks who were starting to develop software tools.
Things didn't exist, so we had to develop them from scratch. This was right around the
time the internet was just becoming, you know, something that we could use as another tool,
too. And so it just evolved over time, and I had the wonderful opportunity almost 13 years ago now
(35:49):
to come back to the NIH, this time at the NCI, and now apply all of that life experience from,
you know, biochemistry to molecular biology to IT and informatics into this role at the NCI. So
it's been a number of doors that have opened at very opportune times. And I just sort of,
(36:12):
you know, I looked for the open doors and I ran through them.
OLIVER BOGLER (36:16):
So you both talked about seeking a way to make greater impact and also just staying
with a field, watching a field that's exploding and evolving and just being part of that. Jeff,
Tony told us how he got back to NCI. I wonder if
you could complete your story. What led you to the current role that you're in?
JEFF SHILLING (36:35):
Well, you know, as Tony was speaking, I thought I wanted to make sure
to communicate (36:40):
there's no clear path to what we're doing that I understand.
I think maybe you get a degree in data science or, you know, you could certainly
get degrees in information technology today, but what we have is this mixture.
I wanted to say this going backwards a little bit is, so I was in the lab and I needed to get a job
(37:02):
in IT. Well, I don't have any IT experience at all other than just doing a few computers in the lab.
I called this woman, Lisa Longenecker, and every day, about this open position, for two months till
she finally hired me half time. And then I had to find another half time job. So for two years,
(37:24):
I worked half time in two departments just to learn. And I was so focused, like Tony said,
you're jumping at every opportunity to learn because I didn't have any formal training. So
then my wife took a postdoc at National Institute of Environmental Health Sciences and I was able
to get a job there. And then after a few years switched from just doing IT support to doing
(37:50):
more IT programming and general IT work at the National Toxicology Program, which is associated
with NIEHS as well. So then a couple of years, we went to Lawrence Livermore Laboratory. And then
my wife and I were both recruited here in the intramural program, Center for Cancer Research.
(38:11):
I remember very much because my wife was interviewing for a job and I just happened
to go with her. So we're at dinner. And we knew the director had worked at NIEHS,
Carl Barrett. And as he was interviewing her, he said, Jeff, what do you do? And I said,
I do this. And he's like, oh my God, I need you more than Michelle. So really,
again, you can call it a door. You can call it whatever. But I think the vacuum of people
(38:37):
who do what we do is still great. If you just get these skills, just like Tony said,
he's pulling wire. He doesn't have to do that. He could have sat back and said,
we can't do this. We don't have any wire. He's like, well, how hard can wire be to pull?
He probably got yelled at. I guarantee he got yelled at for doing that. But we need to do
(38:58):
this. The show must go on. Say however you want to say it. So really, that was it. I think it was
having the skill set, being able to articulate how you can connect the science and the IT or
the informatics, how you can do that. And build the relationships along the way. You have to have
these relationships with the scientists to be able to, I think, get some of these positions.
OLIVER BOGLER (39:23):
Sounds like, almost like a startup ethos, right? That you have to bring to this,
to work in large complex organizations.
JEFF SHILLING (39:30):
Yeah, I wouldn't disagree. I can only say I've never done a startup,
unfortunately. It sounds like fun.
OLIVER BOGLER (39:37):
So in closing then, I wonder what advice you might have to our listeners,
people who are interested in data science and cancer research. Obviously they may have some
more opportunities that didn't exist when you were at their stage in terms of structured
training opportunities. But Tony, let's start with you. What would you say to them?
TONY KERLAVAGE (39:55):
I, you know, again, building on, you know,
both the path that Jeff and I both took, the take home lesson, I think, is flexibility,
is being open to possibilities. You never know what path a career is going to take. I think
back to when I was working at the Institute for Genomic Research, and we were hiring,
(40:20):
you know, it wasn't just molecular biologists and biochemists, we were hiring physicists.
We're hiring people from various walks of life coming together with different training,
different backgrounds and skills coming together to solve a big problem. So my advice would be,
don't think that you don't have the right background, the right training to get into
(40:42):
something that you might have a passion for. Just take that leap of faith to go
through that door and it could be a wonderful opportunity for you that you never even imagined.
OLIVER BOGLER (40:54):
Jeff?
JEFF SHILLING (40:55):
I would say one of my sayings is, when you're an expert, everything you know is old.
No one knows the future. No one knows tomorrow. There is an opportunity for you to learn that.
You have to understand that when you go into a new field, you're going to be overconfident
just because you read a few things that you know what's going on. But the reality is the
(41:19):
situation is very complex. That's why they need people to help figure it out. And so,
you know, just going there, humble, understand that what you know from your other work is very,
very valuable and that learning these new fields is - everyone's in the same
boat. No one knows what's going on. We’re all trying to stay afloat and continuously learn.
(41:43):
And you will one day be the expert, and everything you know will be old.
OLIVER BOGLER (41:48):
Well, thank you both very much for this very
interesting conversation and sharing your advice.
TONY KERLAVAGE (41:53):
You're quite welcome, it's been a pleasure.
JEFF SHILLING (41:56):
Thank you, Oliver.
Oliver Bogler
Now it's time for a segment we call your turn because it's a chance for
our listeners to send in recommendations that they would like to share. If you're listening,
then you're invited to take your turn. Send us a tip for a book, a video, a podcast,
or a talk that you found inspirational, amusing, interesting. You can send these to us
(42:20):
at NCIICC@nih.gov. Record a voice memo and send it along. We may just play it in an upcoming episode.
Now I'd like to invite our guests to take their turn. Let's start with you, Tony.
TONY KERLAVAGE (42:34):
Oliver, there's a book I read recently that I found very impactful. And the
book is called Braiding Sweetgrass by Robin Wall Kimmerer. The subtitle of the book is Indigenous
Wisdom, Scientific Knowledge, and the Teaching of Plants. Robin Wall Kimmerer is a botanist of
Native American heritage and a mother just trying to pass on wisdom to her children.
(43:00):
What I found most fascinating about this book is that what came through was really a new way
of thinking about nature and our place in it. And I highly recommend that book for anyone.
OLIVER BOGLER (43:12):
That sounds interesting, I'm gonna put that on my reading list, thanks. Jeff?
JEFF SHILLING (43:16):
Well, you know, I've always been somewhat of an athlete. I don't really
necessarily always like to watch sports, but I do like to participate in it. And as one gets older,
it's a little harder to do things. So I've, as an adult, I've always really liked to play
golf. And I do read a lot of golf books, both physically and mentally. And one of
the books just came out is called Golf Beneath the Surface: The New Science of Golf Psychology.
(43:44):
And when I started to read it, I was blown away by how very,
very different it was. And it was utilizing all the work that's been done on the brain.
OLIVER BOGLER (43:56):
Hmm, interesting.
JEFF SHILLING (43:57):
And really describing, you know, I was like, this is not a book about
golf at all. It's about how we think and act. How, what do we do when we get afraid? How to
manage our nervousness? What kind of nervousness is good, and is acceptable and how to deal with
that? What kind of nervousness or fear is bad? And that's a sign that we need to change our thinking.
(44:22):
I was so surprised how I'd never heard of these things. Sometimes
I think we have little techniques that we do to manage our life and someone makes
us angry or we're in traffic and we're mad or something like that. I was very,
very surprised and also very excited because all of a sudden people are using all that brain
(44:44):
research that's been happening. And to me, this is kind of the transformation of society to say,
you know, this way we feel, these emotions that we have, that's very, very natural. That is not
something you need to even be concerned about. We were not trained in any of those things, right?
OLIVER BOGLER (45:04):
Sounds useful even if you don't play golf, I guess.
JEFF SHILLING (45:06):
Very useful, yes.
OLIVER BOGLER (45:08):
OK, finally, I'd like to make a recommendation as well. It's a favorite book
of mine that I recently reread. It's called The Diamond Age or A Young Lady's Illustrated Primer
by Neal Stephenson. It was first published in 2000. And I'd remembered it as a book with a
striking vision of how nanotechnology could transform our world when I read it back then.
(45:28):
But this time I enjoyed it much more as a vision of the primer itself. It's a book
that teaches a young girl and imprints on her to provide a bespoke interactive
experience driven by an AI. And it struck me as an incredibly interesting imagining
of what interactive learning might look like with the AI tech that we've
been talking about today. Remarkable book, I think, and well worth a read.
(45:53):
That’s all we have time for on today’s episode of Inside Cancer
Careers! Thank you for joining and thank you to our guests.
We want to hear from you – your stories, your ideas and your feedback are always
welcome. And you are invited to take your turn and make a recommendation
to share with our listeners. You can reach us at NCIICC@nih.gov.
(46:14):
Inside Cancer Careers is a collaboration between NCI’s Office of Communications and
Public Liaison and the Center for Cancer Training.
It is produced by Angela Jones and Astrid Masfar.
Join us every first and third Thursday of the month when new episodes can be found wherever you
listen – subscribe so you won’t miss an episode. I'm your host Oliver Bogler from the National
(46:37):
Cancer Institute and I look forward to sharing your stories here on Inside Cancer Careers.
If you have questions about cancer or comments about this podcast, email us at NCIinfo@nih.gov
or call us at 800-422-6237. And please be sure to mention Inside Cancer Careers in your query.
(47:00):
We are a production of the U.S. Department of Health and Human Services,
National Institutes of Health, National Cancer Institute. Thanks for listening.
Popular Podcasts
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.